"Super Intelligence" replaces "AI" in US federal documents: behind the rebranding, a real political question
🔎 Vocabulary as a political weapon
On September 29, 2026, Donald Trump signs an executive order that changes no line of code, no budget, no regulation. It changes the words. "Artificial Intelligence" becomes "Super Intelligence" throughout the federal executive branch. A week later, on Truth Social, he declares "THE ENEMY" anyone who refuses the new term. Silicon Valley falls in line: Musk renames SpaceXAI to SpaceXSI, Benioff turns AIForce into SIForce, Nadella titles his essay on the emergency brake "Super Intelligence". This is no joke. It's a semantic capture operation at industrial scale, at the precise moment when Europe sends AI Act cease-and-desist letters to 30 companies and the White House creates a "Super Intelligence Force" tasked with requiring incident disclosure. The word "superintelligence" referred, in Bostrom's work and in the research community, to a hypothetical AI surpassing humanity as a whole. The executive order applies it to AI, plain and simple. The blurring is total. And that's the point.
Key takeaways
- Executive order of September 29, 2026: "Inaugurating the Era of Super Intelligence" mandates the term "Super Intelligence" (SI) in all correspondence, policy documents, and public communications of the federal executive branch →.
- No legal changes: Section 2 prohibits modifying regulations, contracts, grants, or historical documents; Section 4 specifies that no actionable right is created →.
- Federal definition within 60 days: the OSTP (Advisor to the President for Science and Technology) must propose a legislative definition by the end of November 2026, by reference to the existing AI definition (15 USC §9401(3)) →.
- Voluntary "White House Accord on Super Intelligence": signed on September 29 with six leaders (Google, Meta, Anthropic, Nvidia, OpenAI, Musk) — zero sanctions provided for →.
- Tech buy-in: Musk (SpaceXSI), Benioff (SIForce), Nadella (trying out "Super Intelligence"), and Nvidia adopt the vocabulary between October 8 and 12 →→.
- Parallel regulatory context: EU — AI Act letters to 30 companies; US — the "White House Super Intelligence Force" requires incident disclosure →.
Recommended Tools
| Tool | Main use | Price (October 2026) | Ideal for |
|---|---|---|---|
| Hostinger | Hosting + .si domain | €2.99/month (October 2026, check on hostinger.com) | Securing a .si domain before the rush |
| Gemini 3.1 Pro (Google) | Regulatory document analysis | Included in Google AI Studio (free with quotas) | Reading the decree and the AI Act side by side |
| GPT-5.5 (OpenAI) | Multilingual regulatory monitoring | $20/month (Plus) / $200/month (Pro) | Synthesizing US + EU sources |
| Claude Opus 4.7 (Anthropic) | Assisted legal drafting | $20/month (Pro) | Writing SI/AI Act compliance memos |
To decide between these assistants based on your use case (monitoring, document analysis, drafting), check out our comparison Google Gemini vs ChatGPT vs Claude: which one for which use?.
The September 29 executive order — what really changes, and what doesn't
The direct answer: almost nothing on the legal front, everything on the symbolic and communications front.
The executive order "Inaugurating the Era of Super Intelligence" signed on September 29, 2026 is a three-page text. Its Section 1 orders all federal agencies to use "Super Intelligence" and "SI" in official correspondence, public communications, policy documents, and non-legislative documents, and to "no longer recognize" the terms "Artificial Intelligence" or "AI" →. Its Section 2, the crucial one, specifies that "nothing in this order shall be construed as modifying any regulation, presidential action, contract, grant, or historical document." Its Section 4 adds that it "creates no right enforceable in court." Its Section 3 defines "Super Intelligence" by simple cross-reference to the legal definition of "artificial intelligence" in section 9401(3) of title 15 of the United States Code — the 2020 federal law →.
In short: the term changes, the legal definition stays the same, no rule moves, and no one can challenge the government on this basis. It is a performative order: it brings a reality into being simply by naming it. Trump owns it: "The word super is the best of all, and it's the simplest," he says, adding that Xi Jinping, whom he had hosted the previous week, would share that taste →.
The same day, a "White House Accord on Super Intelligence" is signed over lunch with six CEOs: Sundar Pichai (Google), Mark Zuckerberg (Meta), Dario Amodei (Anthropic), Jensen Huang (Nvidia), Sam Altman (OpenAI), Elon Musk (xAI). A voluntary accord, no sanctions, no quantified commitments. A family photo with a new logo →.
The timeline — from the UN to Truth Social, the escalation in three weeks
The direct answer: a calibrated staging, from the UN speech to the executive order, then to the public injunction of "THE ENEMY."
- September 22, 2026: Trump speaks at the UN General Assembly. An oral statement with no published text: he evokes a "new era of Super Intelligence" without signing any document →.
- September 29, 2026: Signing of the executive order "Inaugurating the Era of Super Intelligence" at the White House. Lunch with the six tech leaders. Signing of the Voluntary Agreement. Publication of the Fact Sheet and the full text that same evening →.
- October 8, 2026: Truth Social. Trump writes in capital letters: "The White House considers THE ENEMY anyone who uses the term 'Artificial Intelligence' instead of the new, very accepted, and more precise term 'Super Intelligence'!" →. Politico notes that Trump regularly describes AI surveillance whistleblowers as "exaggerated" and a "hoax" →.
- October 12, 2026: The Verge (via TEXXR) takes stock of the alignments: Musk announced SpaceXAI → SpaceXSI, Benioff AIForce → SIForce, Nadella and Nvidia adopt the term — Nadella in the very title of his essay on the emergency brake →. Business Insider confirms: "The era of Super Intelligence is here. AIForce is officially SIForce" →.
Three weeks. An escalation from symbol to injunction, from executive order to public pressure on private vocabulary. The timeline is no accident: it coincides with the European Commission sending AI Act letters of formal notice to some thirty companies (late September – early October), and with the quiet establishment of a "White House Super Intelligence Force" tasked with collecting incident reports →.
Semantic Drift — Bostrom vs. Trump, or How to Empty a Concept of Its Substance
The direct answer: "Superintelligence" referred to a future AI surpassing humanity; the executive order applies it to today's AI. The debate over guardrails comes out of it muddied.
Nick Bostrom, in Superintelligence (2014), defines the term as "any intellect that greatly outperforms the cognitive performance of humans in practically all domains of interest." It is a forward-looking, speculative notion, tied to existential risk. The research community (Future of Humanity Institute, MIRI, OpenAI in its early days) uses it to distinguish AGI (general intelligence) from what would come after — a recursive, self-improving, potentially uncontrollable intelligence.
The executive order of September 29, 2026 takes this term of art and applies it to any AI covered by the 2020 legal definition (15 USC §9401(3)): "a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments" →. ChatGPT, Gemini, a Netflix recommendation system, a spam filter: everything becomes "Super Intelligence."
OIA.fr sums it up bluntly: "In research vocabulary, 'superintelligence' refers to a hypothetical AI more capable than the most capable humans, one that does not yet exist; the order uses it for AI plain and simple, which muddies the guardrails debate at a time when part of the industry is calling for a slowdown" →.
This blurring is no small matter. When the safety vocabulary (alignment, interpretability, scalable oversight) rests on a distinction between "current AI" and "future superintelligence," erasing the linguistic boundary erases the conceptual one. Regulators lose a reference point. Journalists lose a nuance. The public loses a cognitive guardrail. And the industry gains room to maneuver: no need to promise prudence for a "superintelligence" that doesn't exist — we're already in it, by decree.
Silicon Valley's adoption — marketing, allegiance, or both?
The direct answer: both. The rebranding provides political cover for commercial renamings that were already ready to go; refusing would carry an immediate political cost.
Elon Musk is the first to move. On Sunday, October 4 (according to Business Insider), he promises to rename SpaceXAI to SpaceXSI →. On Friday, October 9, Marc Benioff announces that AIForce becomes SIForce: "The era of Super Intelligence is here" →. On October 12, The Verge notes that Satya Nadella and Nvidia have adopted the term — Nadella in the title of his essay calling for an "emergency brake" on AI →.
These moves are not spontaneous. SpaceXAI, AIForce: these brands already existed. The decree hands them a window of opportunity for a "patriotic" rebranding that costs nothing in R&D and pays off in visibility. But there's more: on October 8, Trump designated as "THE ENEMY" anyone who refused the term. For a publicly traded company whose federal contracts run into the billions (Microsoft, Nvidia, Google via Cloud), the cost of symbolic disobedience is prohibitive. Nadella cannot afford to be the president's "THE ENEMY" when Azure powers entire swaths of the administration.
The anecdote reported by The Verge speaks volumes: "administration insiders reportedly bought up .si domains en masse from the very start" →. Speculation on the .si extension (Slovenia) reveals the cynical anticipation of a terminology bubble. Vocabulary becomes a financial asset.
The regulatory parallel — White House Super Intelligence Force vs the European AI Act
The direct answer: two opposing approaches: Europe regulates through binding law; the US regulates through vocabulary and voluntary incident disclosure.
On the European side: late September - early October 2026, the European Commission sends formal notice letters to some thirty companies for non-compliance with the AI Act — in force since August 2024, with the provisions on general-purpose models applicable since August 2025. Sanctions: up to 3% of global revenue or 15 million euros →. The approach is legal, graduated, transparent, with technical thresholds (FLOPs, data, capabilities).
On the American side: the "White House Super Intelligence Force" (created in the wake of the executive order, with no distinct public text) requires incident disclosure from the labs that signed the voluntary Accord. No sanctions. No public technical thresholds. An email inbox, reputational pressure, an imposed vocabulary →.
The contrast is striking. Europe names things (unacceptable-risk AI, high-risk AI, general-purpose AI) in order to regulate them. The US renames everything "Super Intelligence" in order to avoid regulating — the executive order changes no rules, the Accord commits to nothing, the Force has no teeth. But the two approaches converge on one point: the demand for transparency about incidents. The AI Act mandates it (Article 73). The Super Intelligence Force calls for it. That is the only common ground.
For companies operating on both sides of the Atlantic, the terminological schizophrenia is a real compliance cost. The same model is a "general-purpose AI with systemic risk" in Brussels, "Super Intelligence" in Washington. Technical documentation, incident registers, public communications: everything must be duplicated, adapted, translated. Not in language — in regime of truth.
What the executive order doesn't change — and why that matters more than what it does
The direct answer: the executive order is legally an empty shell, but a loaded political weapon. Its inaction is its message.
Let's reread sections 2 and 4. Section 2: "Nothing in this order shall be construed as modifying any regulation, presidential action, contract, grant, or historical document." Section 4: "This order creates no right enforceable in a court of law." →.
Concretely, this means:
- Federal contracts signed yesterday with "AI" remain valid.
- Existing regulations (previous executive orders, OMB memos, NIST AI RMF) keep their "AI" vocabulary.
- NSF, DOE, DARPA grants continue to use "AI" in their calls for proposals.
- No citizen, no NGO, no company can go to court to force the administration to say "Super Intelligence" — or to stop it from doing so.
So what's the point? Three things. First: locking in the media narrative. Every article, every press release, every congressional testimony will now have to use "SI" or risk being labeled an "ENEMY". Second: preparing the legislative ground. The APST has 60 days (until November 29, 2026) to propose a federal definition of "Super Intelligence" with "conforming amendments" — that is, rewriting existing laws to insert the new term →. Third: creating a loyalty marker. Saying "AI" becomes an act of resistance; saying "SI" becomes an act of allegiance. This is politics by dictionary.
The Implications for Businesses and Developers — Compliance, Communication, Monitoring
The direct answer: no mandatory technical changes, but an immediate communications compliance cost and a risk of transatlantic regulatory misalignment.
For SaaS / AI Software Vendors
Your federal contracts (FedRAMP, GSA Schedule, DoD contracts) are not changing. Your technical documentation (API docs, model cards, system cards) does not legally have to be rewritten. But your public communications, blog posts, press releases, Congressional testimony: if you sell to the federal government, the "SI" vocabulary becomes the expected norm. Refusing to adopt it exposes you to being politically labeled an "ENEMY" — a reputational risk few CEOs will be willing to bear.
For Legal / Compliance Teams
Two parallel regimes of truth to manage:
- USA: definition by cross-reference (15 USC §9401(3)), "SI" vocabulary mandated across the executive branch, voluntary incident disclosure via Super Intelligence Force.
- EU: technical definition (AI Act, Annex III + FLOPs thresholds), "general-purpose AI / systemic-risk AI" vocabulary, mandatory incident disclosure (art. 73), financial penalties.
Your incident registers, your impact assessments, your transparency documentation: you need two versions. Not a translation — a structural adaptation. The same incident can be "notifiable" in Brussels (severity threshold) and "non-reportable" in Washington (no threshold, voluntary only).
For Developers / Data Scientists
Your models, your pipelines, your metrics: unchanged. But your model cards (model documentation) intended for US public-sector buyers will have to mention "Super Intelligence" somewhere. Your internal benchmarks: watch out for academic papers that cite "superintelligence" in the Bostrom sense — cite them with a footnote: "In the research sense, not in the sense of the 2026 US executive order." Otherwise, confusion is guaranteed.
Toward a federal definition — the 60-day countdown
The direct answer: by November 29, 2026, the OSTP must deliver a legislative proposal. That will be the real test: a technical definition or a political definition?
The executive order gives the OSTP's Acting Director (currently Michael Kratsios, according to sources) 60 days to "propose a federal definition of 'Super Intelligence' with conforming amendments" →. "Conforming amendments": a legal term meaning "changes to existing laws to replace 'AI' with 'Super Intelligence' and ensure consistency."
Three scenarios:
- Definition by pure cross-reference: "Super Intelligence = artificial intelligence within the meaning of 15 USC §9401(3)". Result: zero substantive change, purely symbolic. The legislative status quo with a fresh coat of paint.
- Definition by technical thresholds: FLOPs, parameters, benchmarks (MMLU, GPQA, SWE-bench), agentic capabilities. Result: possible alignment with the European AI Act (10^25 FLOPs threshold for "systemic risk"), but a break with the historical US approach (no hard thresholds).
- Broad political definition: "Any system designated as such by the Secretary of Commerce / the Director of the OSTP." Result: discretionary, opaque, instrumentalized.
Scenario 1 is the most likely, given section 3 of the executive order (explicit reference to the 2020 definition). But scenario 3 is the most dangerous: it would give the executive branch control over what counts as Super Intelligence — and therefore over who falls under the Super Intelligence Force, who must disclose their incidents, who is "THE ENEMY" if they refuse the label.
The technical community (NIST, academics, think tanks) has 60 days to produce counter-proposals, impact analyses, citizen amendments. Silence constitutes acceptance.
❌ Common Mistakes
Mistake 1: Believing the executive order changes AI regulation in the US
What goes wrong: Reading the title "Inaugurating the Era of Super Intelligence" and thinking a new binding law has come into being. The fix: Read sections 2 and 4. No regulation changed, no actionable rights. It's a decree about vocabulary, not substance.
Mistake 2: Confusing "Super Intelligence" (US executive order) and "superintelligence" (Bostrom's sense)
What goes wrong: Using the two terms interchangeably in a risk analysis. The fix: Explicitly distinguish: "Super Intelligence (US executive order, = AI in the broad sense)" vs "superintelligence (research, = AI surpassing humanity, does not yet exist)". A footnote is mandatory.
Mistake 3: Neglecting the transatlantic compliance cost
What goes wrong: Producing a single transparency sheet / incident register for both markets. The fix: Two distinct artifacts, mapped onto two definitional regimes and two notification thresholds. Automate the dual generation.
Mistake 4: Underestimating the "THE ENEMY" effect on the supply chain
What goes wrong: Thinking only the big labs (OpenAI, Anthropic, Google) are affected. The fix: Every subcontractor, data provider, and component vendor selling to the US federal government must audit its contractual and marketing vocabulary. The decree trickles down through capillary action.
❓ Frequently Asked Questions
Does the executive order apply to private companies that don't work with the federal government?
No. The order binds only "the federal executive" (agencies, departments). But political pressure ("THE ENEMY") and the contractual cascade effect (subcontractors of subcontractors) create a de facto standard.
Has the legal definition of AI (15 USC §9401(3)) changed?
No. Section 3 of the order explicitly refers to it. The 2020 definition still stands: "a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments."
Does the voluntary "White House Accord on Super Intelligence" have force of law?
No. Zero sanctions, zero quantified commitments, zero verification mechanism. It's a declaration of intent signed by six CEOs to accompany the presidential lunch.
Will Europe adopt the term "Super Intelligence"?
No sign of it. The AI Act uses "AI system", "general-purpose AI model", "systemic-risk AI". The European vocabulary is technical, not branding. Indeed, the Commission sent cease-and-desist letters to 30 companies under the current vocabulary in late September 2026.
What becomes of the .si domains bought by "insiders"?
Pure speculation. The .si extension (Slovenia) has no official connection to "Super Intelligence". But the rumor (The Verge via TEXXR) shows anticipation of a terminology bubble →.
Did Nadella really use "Super Intelligence" in the title of his essay?
Yes. The Verge (October 12) confirms that Nadella's essay calling for an "emergency brake" on AI uses "Super Intelligence" in its title →. This is the most significant endorsement yet: Microsoft is the leading cloud provider to the US federal government.
✅ Conclusion
The "Super Intelligence" executive order doesn't regulate — it marks. It plants a terminological flag on the territory of American AI, forces Silicon Valley to salute the flag, and prepares a legislative rewrite whose substance (technical or political) will be decided by November 29. Meanwhile, Europe is regulating for real. Companies caught between the two have no choice: two vocabularies, two registers, two logics. "Super Intelligence" is not a model. It's a power play. →
To dig deeper: keep up with transatlantic regulatory monitoring through our analyses of the federal evaluation of the 5 major US labs (CAISI) and the Microsoft Build 2026 strategy around MAI-Thinking. On the infrastructure side, Google Gemini Intelligence turns Android into an "intelligence system" — the vocabulary follows the hardware.
```